A two-column paper can produce perfectly recognizable words in the wrong sequence. The extracted text may start in the left column, jump to a caption on the right, and then return to finish the sentence. Correcting spelling will not repair that argument.
Check reading order as a relationship between blocks. You need the original PDF beside the conversion result, not just a spellchecker and a word count.
Draw a route through one page
Pick a page containing two ordinary columns, a heading, and at least one figure or table. Read it in the order the author intended. For a conventional left-to-right paper, that often means down the left column and then down the right, but full-width headings and figures can interrupt that pattern.
Write a short route in your notes, such as:
Section heading → upper left paragraph → lower left paragraph → upper right continuation → next section.
This is your review map, not a claim about how every PDF is encoded. Some pages switch between one and two columns. An abstract may span the page even when the following sections do not.
In the converted text, find the first few words of each mapped paragraph. Confirm that the sequence matches. If it does not, identify the earliest point where the two versions diverge rather than moving paragraphs at random.
Treat full-width headings as boundaries
A heading across both columns often introduces a new reading region. Check whether the result places it before both columns, halfway through the left column, or after text that actually belongs below it.
Compare the paragraph before the heading and the first paragraph after it. A heading can appear correct in isolation while separating the wrong two blocks. Running titles at the top of successive pages are another trap: the extractor may treat them as repeated section headings.
For a paper with an inset box, distinguish the main argument from the box's separate text. Do not insert a definition or sidebar into the middle of an unrelated sentence simply because their vertical positions overlap.
Check where figures and tables enter the text
Follow a reference such as “see Figure 3” to the actual illustration and its caption. The caption should stay associated with that figure, even if its position changes in a flowing output.
A wide table may sit between the two halves of a page. Inspect whether the text that follows it continues the first column, starts a new subsection, or resumes the second column. A conversion can keep every table value and still place the entire table inside a paragraph.
For each figure you sample, compare:
- Its number and the reference in the body.
- The start and end of the caption.
- Any labels needed to interpret it.
- The paragraph immediately before and after the insertion.
Do not use a picture's physical proximity as the only test. Authors sometimes discuss a figure before or after the page where it is printed.
Read across transitions, not just inside paragraphs
Most ordering faults become visible at boundaries. Read the last two sentences at the bottom of the left column and the first two at the top of the right. Then do the same across a page break.
Watch for a sudden change of subject, an undefined pronoun, an equation without its introduction, or a caption embedded in a sentence. A hyphen at the end of a printed line may join a word, but it may also belong to a real compound. Restore it only after checking the source.
Keep a compact record: original PDF page, paragraph opening, incorrect neighbor, intended neighbor, and proposed correction. This lets you distinguish a local misplaced caption from a repeated ordering problem across the paper. The Markdown reading-order troubleshooting guide starts from those output symptoms.
Choose Markdown for repair or EPUB for reading
PDF to Markdown is useful when you need to move blocks, restore headings, or collect research notes in an editor. Upload the PDF, select Markdown in task settings, and submit the conversion. PDF Craft processes layout and recognition, including multi-column material, but complex pages still need inspection.
Choose EPUB when sustained reading and adjustable text are the main goal. Reflow changes positions, so matching printed page numbers is a poor acceptance rule. Compare content order instead. Keep the original PDF for diagrams and citations whichever output you choose.
Before importing the result into a notes collection, review more than a clean first page. Sample a full-width heading, a dense two-column page, a wide table, and the end of the document. Then follow the research Markdown workflow to preserve the source alongside your notes.